AI Update Deep Dive: Synthetic Images Against Bias

▤ Summary

The article explains how biased training data can reduce the reliability of medical AI, especially in melanoma detection. It highlights MedGenAI, a project that uses generative AI to create synthetic skin disease images across controlled parameters such as age, gender, and skin type. The team also uses counterfactual tests to measure whether models change their predictions when only one attribute shifts. The goal is to expose and reduce demographic bias before any clinical deployment. The piece stresses that synthetic data complements real patient data and still requires expert review and clinical studies.

▦ Linked Companies